1 // SPDX-License-Identifier: GPL-2.0
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
16 #include "btree_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
28 #include "disk_groups.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
40 #include "journal_reclaim.h"
41 #include "journal_seq_blacklist.h"
45 #include "nocow_locking.h"
47 #include "rebalance.h"
52 #include "subvolume.h"
58 #include <linux/backing-dev.h>
59 #include <linux/blkdev.h>
60 #include <linux/debugfs.h>
61 #include <linux/device.h>
62 #include <linux/idr.h>
63 #include <linux/module.h>
64 #include <linux/percpu.h>
65 #include <linux/random.h>
66 #include <linux/sysfs.h>
67 #include <crypto/hash.h>
69 MODULE_LICENSE("GPL");
70 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
73 static const struct attribute_group type ## _group = { \
74 .attrs = type ## _files \
77 static const struct attribute_group *type ## _groups[] = { \
82 static const struct kobj_type type ## _ktype = { \
83 .release = type ## _release, \
84 .sysfs_ops = &type ## _sysfs_ops, \
85 .default_groups = type ## _groups \
88 static void bch2_fs_release(struct kobject *);
89 static void bch2_dev_release(struct kobject *);
90 static void bch2_fs_counters_release(struct kobject *k)
94 static void bch2_fs_internal_release(struct kobject *k)
98 static void bch2_fs_opts_dir_release(struct kobject *k)
102 static void bch2_fs_time_stats_release(struct kobject *k)
107 KTYPE(bch2_fs_counters);
108 KTYPE(bch2_fs_internal);
109 KTYPE(bch2_fs_opts_dir);
110 KTYPE(bch2_fs_time_stats);
113 static struct kset *bcachefs_kset;
114 static LIST_HEAD(bch_fs_list);
115 static DEFINE_MUTEX(bch_fs_list_lock);
117 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
119 static void bch2_dev_free(struct bch_dev *);
120 static int bch2_dev_alloc(struct bch_fs *, unsigned);
121 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
122 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
124 struct bch_fs *bch2_dev_to_fs(dev_t dev)
130 mutex_lock(&bch_fs_list_lock);
133 list_for_each_entry(c, &bch_fs_list, list)
134 for_each_member_device_rcu(ca, c, i, NULL)
135 if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
142 mutex_unlock(&bch_fs_list_lock);
147 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
151 lockdep_assert_held(&bch_fs_list_lock);
153 list_for_each_entry(c, &bch_fs_list, list)
154 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
160 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
164 mutex_lock(&bch_fs_list_lock);
165 c = __bch2_uuid_to_fs(uuid);
168 mutex_unlock(&bch_fs_list_lock);
173 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
176 unsigned i, nr = 0, u64s =
177 ((sizeof(struct jset_entry_dev_usage) +
178 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
182 for_each_member_device_rcu(ca, c, i, NULL)
186 bch2_journal_entry_res_resize(&c->journal,
187 &c->dev_usage_journal_res, u64s * nr);
190 /* Filesystem RO/RW: */
193 * For startup/shutdown of RW stuff, the dependencies are:
195 * - foreground writes depend on copygc and rebalance (to free up space)
197 * - copygc and rebalance depend on mark and sweep gc (they actually probably
198 * don't because they either reserve ahead of time or don't block if
199 * allocations fail, but allocations can require mark and sweep gc to run
200 * because of generation number wraparound)
202 * - all of the above depends on the allocator threads
204 * - allocator depends on the journal (when it rewrites prios and gens)
207 static void __bch2_fs_read_only(struct bch_fs *c)
210 unsigned i, clean_passes = 0;
214 bch2_open_buckets_stop(c, NULL, true);
215 bch2_rebalance_stop(c);
217 bch2_gc_thread_stop(c);
220 bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
221 journal_cur_seq(&c->journal));
226 if (bch2_btree_interior_updates_flush(c) ||
227 bch2_journal_flush_all_pins(&c->journal) ||
228 bch2_btree_flush_all_writes(c) ||
229 seq != atomic64_read(&c->journal.seq)) {
230 seq = atomic64_read(&c->journal.seq);
233 } while (clean_passes < 2);
235 bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
236 journal_cur_seq(&c->journal));
238 if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
239 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
240 set_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
241 bch2_fs_journal_stop(&c->journal);
244 * After stopping journal:
246 for_each_member_device(ca, c, i)
247 bch2_dev_allocator_remove(c, ca);
250 #ifndef BCH_WRITE_REF_DEBUG
251 static void bch2_writes_disabled(struct percpu_ref *writes)
253 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
255 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
256 wake_up(&bch2_read_only_wait);
260 void bch2_fs_read_only(struct bch_fs *c)
262 if (!test_bit(BCH_FS_RW, &c->flags)) {
263 bch2_journal_reclaim_stop(&c->journal);
267 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
270 * Block new foreground-end write operations from starting - any new
271 * writes will return -EROFS:
273 set_bit(BCH_FS_GOING_RO, &c->flags);
274 #ifndef BCH_WRITE_REF_DEBUG
275 percpu_ref_kill(&c->writes);
277 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
278 bch2_write_ref_put(c, i);
282 * If we're not doing an emergency shutdown, we want to wait on
283 * outstanding writes to complete so they don't see spurious errors due
284 * to shutting down the allocator:
286 * If we are doing an emergency shutdown outstanding writes may
287 * hang until we shutdown the allocator so we don't want to wait
288 * on outstanding writes before shutting everything down - but
289 * we do need to wait on them before returning and signalling
290 * that going RO is complete:
292 wait_event(bch2_read_only_wait,
293 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
294 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
296 __bch2_fs_read_only(c);
298 wait_event(bch2_read_only_wait,
299 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
301 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
302 clear_bit(BCH_FS_GOING_RO, &c->flags);
304 if (!bch2_journal_error(&c->journal) &&
305 !test_bit(BCH_FS_ERROR, &c->flags) &&
306 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
307 test_bit(BCH_FS_STARTED, &c->flags) &&
308 test_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags) &&
309 !c->opts.norecovery) {
310 BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
311 BUG_ON(atomic_read(&c->btree_cache.dirty));
312 BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
313 BUG_ON(c->btree_write_buffer.state.nr);
315 bch_verbose(c, "marking filesystem clean");
316 bch2_fs_mark_clean(c);
319 clear_bit(BCH_FS_RW, &c->flags);
322 static void bch2_fs_read_only_work(struct work_struct *work)
325 container_of(work, struct bch_fs, read_only_work);
327 down_write(&c->state_lock);
328 bch2_fs_read_only(c);
329 up_write(&c->state_lock);
332 static void bch2_fs_read_only_async(struct bch_fs *c)
334 queue_work(system_long_wq, &c->read_only_work);
337 bool bch2_fs_emergency_read_only(struct bch_fs *c)
339 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
341 bch2_journal_halt(&c->journal);
342 bch2_fs_read_only_async(c);
344 wake_up(&bch2_read_only_wait);
348 static int bch2_fs_read_write_late(struct bch_fs *c)
353 * Data move operations can't run until after check_snapshots has
354 * completed, and bch2_snapshot_is_ancestor() is available.
356 * Ideally we'd start copygc/rebalance earlier instead of waiting for
357 * all of recovery/fsck to complete:
359 ret = bch2_copygc_start(c);
361 bch_err(c, "error starting copygc thread");
365 ret = bch2_rebalance_start(c);
367 bch_err(c, "error starting rebalance thread");
374 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
380 if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
381 bch_err(c, "cannot go rw, unfixed btree errors");
382 return -BCH_ERR_erofs_unfixed_errors;
385 if (test_bit(BCH_FS_RW, &c->flags))
388 if (c->opts.norecovery)
389 return -BCH_ERR_erofs_norecovery;
392 * nochanges is used for fsck -n mode - we have to allow going rw
393 * during recovery for that to work:
395 if (c->opts.nochanges && (!early || c->opts.read_only))
396 return -BCH_ERR_erofs_nochanges;
398 bch_info(c, "going read-write");
400 ret = bch2_fs_mark_dirty(c);
404 clear_bit(BCH_FS_CLEAN_SHUTDOWN, &c->flags);
407 * First journal write must be a flush write: after a clean shutdown we
408 * don't read the journal, so the first journal write may end up
409 * overwriting whatever was there previously, and there must always be
410 * at least one non-flush write in the journal or recovery will fail:
412 set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
414 for_each_rw_member(ca, c, i)
415 bch2_dev_allocator_add(c, ca);
416 bch2_recalc_capacity(c);
418 ret = bch2_gc_thread_start(c);
420 bch_err(c, "error starting gc thread");
425 ret = bch2_fs_read_write_late(c);
430 #ifndef BCH_WRITE_REF_DEBUG
431 percpu_ref_reinit(&c->writes);
433 for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
434 BUG_ON(atomic_long_read(&c->writes[i]));
435 atomic_long_inc(&c->writes[i]);
438 set_bit(BCH_FS_RW, &c->flags);
439 set_bit(BCH_FS_WAS_RW, &c->flags);
442 bch2_do_invalidates(c);
443 bch2_do_stripe_deletes(c);
444 bch2_do_pending_node_rewrites(c);
447 __bch2_fs_read_only(c);
451 int bch2_fs_read_write(struct bch_fs *c)
453 return __bch2_fs_read_write(c, false);
456 int bch2_fs_read_write_early(struct bch_fs *c)
458 lockdep_assert_held(&c->state_lock);
460 return __bch2_fs_read_write(c, true);
463 /* Filesystem startup/shutdown: */
465 static void __bch2_fs_free(struct bch_fs *c)
470 for (i = 0; i < BCH_TIME_STAT_NR; i++)
471 bch2_time_stats_exit(&c->times[i]);
473 bch2_free_pending_node_rewrites(c);
474 bch2_fs_counters_exit(c);
475 bch2_fs_snapshots_exit(c);
476 bch2_fs_quota_exit(c);
477 bch2_fs_fs_io_direct_exit(c);
478 bch2_fs_fs_io_buffered_exit(c);
479 bch2_fs_fsio_exit(c);
481 bch2_fs_encryption_exit(c);
483 bch2_fs_buckets_waiting_for_journal_exit(c);
484 bch2_fs_btree_interior_update_exit(c);
485 bch2_fs_btree_iter_exit(c);
486 bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
487 bch2_fs_btree_cache_exit(c);
488 bch2_fs_replicas_exit(c);
489 bch2_fs_journal_exit(&c->journal);
490 bch2_io_clock_exit(&c->io_clock[WRITE]);
491 bch2_io_clock_exit(&c->io_clock[READ]);
492 bch2_fs_compress_exit(c);
493 bch2_journal_keys_free(&c->journal_keys);
494 bch2_journal_entries_free(c);
495 bch2_fs_btree_write_buffer_exit(c);
496 percpu_free_rwsem(&c->mark_lock);
497 free_percpu(c->online_reserved);
499 if (c->btree_paths_bufs)
500 for_each_possible_cpu(cpu)
501 kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
503 darray_exit(&c->btree_roots_extra);
504 free_percpu(c->btree_paths_bufs);
505 free_percpu(c->pcpu);
506 mempool_exit(&c->large_bkey_pool);
507 mempool_exit(&c->btree_bounce_pool);
508 bioset_exit(&c->btree_bio);
509 mempool_exit(&c->fill_iter);
510 #ifndef BCH_WRITE_REF_DEBUG
511 percpu_ref_exit(&c->writes);
513 kfree(rcu_dereference_protected(c->disk_groups, 1));
514 kfree(c->journal_seq_blacklist_table);
515 kfree(c->unused_inode_hints);
518 destroy_workqueue(c->write_ref_wq);
519 if (c->io_complete_wq)
520 destroy_workqueue(c->io_complete_wq);
522 destroy_workqueue(c->copygc_wq);
523 if (c->btree_io_complete_wq)
524 destroy_workqueue(c->btree_io_complete_wq);
525 if (c->btree_update_wq)
526 destroy_workqueue(c->btree_update_wq);
528 bch2_free_super(&c->disk_sb);
529 kvpfree(c, sizeof(*c));
530 module_put(THIS_MODULE);
533 static void bch2_fs_release(struct kobject *kobj)
535 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
540 void __bch2_fs_stop(struct bch_fs *c)
545 bch_verbose(c, "shutting down");
547 set_bit(BCH_FS_STOPPING, &c->flags);
549 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
551 down_write(&c->state_lock);
552 bch2_fs_read_only(c);
553 up_write(&c->state_lock);
555 for_each_member_device(ca, c, i)
556 if (ca->kobj.state_in_sysfs &&
558 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
560 if (c->kobj.state_in_sysfs)
561 kobject_del(&c->kobj);
563 bch2_fs_debug_exit(c);
564 bch2_fs_chardev_exit(c);
566 kobject_put(&c->counters_kobj);
567 kobject_put(&c->time_stats);
568 kobject_put(&c->opts_dir);
569 kobject_put(&c->internal);
571 /* btree prefetch might have kicked off reads in the background: */
572 bch2_btree_flush_all_reads(c);
574 for_each_member_device(ca, c, i)
575 cancel_work_sync(&ca->io_error_work);
577 cancel_work_sync(&c->read_only_work);
580 void bch2_fs_free(struct bch_fs *c)
584 mutex_lock(&bch_fs_list_lock);
586 mutex_unlock(&bch_fs_list_lock);
588 closure_sync(&c->cl);
589 closure_debug_destroy(&c->cl);
591 for (i = 0; i < c->sb.nr_devices; i++) {
592 struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
595 bch2_free_super(&ca->disk_sb);
600 bch_verbose(c, "shutdown complete");
602 kobject_put(&c->kobj);
605 void bch2_fs_stop(struct bch_fs *c)
611 static int bch2_fs_online(struct bch_fs *c)
617 lockdep_assert_held(&bch_fs_list_lock);
619 if (__bch2_uuid_to_fs(c->sb.uuid)) {
620 bch_err(c, "filesystem UUID already open");
624 ret = bch2_fs_chardev_init(c);
626 bch_err(c, "error creating character device");
630 bch2_fs_debug_init(c);
632 ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
633 kobject_add(&c->internal, &c->kobj, "internal") ?:
634 kobject_add(&c->opts_dir, &c->kobj, "options") ?:
635 kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
636 kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
637 bch2_opts_create_sysfs_files(&c->opts_dir);
639 bch_err(c, "error creating sysfs objects");
643 down_write(&c->state_lock);
645 for_each_member_device(ca, c, i) {
646 ret = bch2_dev_sysfs_online(c, ca);
648 bch_err(c, "error creating sysfs objects");
649 percpu_ref_put(&ca->ref);
654 BUG_ON(!list_empty(&c->list));
655 list_add(&c->list, &bch_fs_list);
657 up_write(&c->state_lock);
661 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
663 struct bch_sb_field_members *mi;
665 struct printbuf name = PRINTBUF;
666 unsigned i, iter_size;
669 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
671 c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
675 __module_get(THIS_MODULE);
677 closure_init(&c->cl, NULL);
679 c->kobj.kset = bcachefs_kset;
680 kobject_init(&c->kobj, &bch2_fs_ktype);
681 kobject_init(&c->internal, &bch2_fs_internal_ktype);
682 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
683 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
684 kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
687 c->disk_sb.fs_sb = true;
689 init_rwsem(&c->state_lock);
690 mutex_init(&c->sb_lock);
691 mutex_init(&c->replicas_gc_lock);
692 mutex_init(&c->btree_root_lock);
693 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
695 init_rwsem(&c->gc_lock);
696 mutex_init(&c->gc_gens_lock);
698 for (i = 0; i < BCH_TIME_STAT_NR; i++)
699 bch2_time_stats_init(&c->times[i]);
701 bch2_fs_copygc_init(c);
702 bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
703 bch2_fs_btree_interior_update_init_early(c);
704 bch2_fs_allocator_background_init(c);
705 bch2_fs_allocator_foreground_init(c);
706 bch2_fs_rebalance_init(c);
707 bch2_fs_quota_init(c);
708 bch2_fs_ec_init_early(c);
709 bch2_fs_move_init(c);
711 INIT_LIST_HEAD(&c->list);
713 mutex_init(&c->usage_scratch_lock);
715 mutex_init(&c->bio_bounce_pages_lock);
716 mutex_init(&c->snapshot_table_lock);
718 spin_lock_init(&c->btree_write_error_lock);
720 INIT_WORK(&c->journal_seq_blacklist_gc_work,
721 bch2_blacklist_entries_gc);
723 INIT_LIST_HEAD(&c->journal_iters);
725 INIT_LIST_HEAD(&c->fsck_errors);
726 mutex_init(&c->fsck_error_lock);
728 seqcount_init(&c->gc_pos_lock);
730 seqcount_init(&c->usage_lock);
732 sema_init(&c->io_in_flight, 128);
734 INIT_LIST_HEAD(&c->vfs_inodes_list);
735 mutex_init(&c->vfs_inodes_lock);
737 c->copy_gc_enabled = 1;
738 c->rebalance.enabled = 1;
739 c->promote_whole_extents = true;
741 c->journal.flush_write_time = &c->times[BCH_TIME_journal_flush_write];
742 c->journal.noflush_write_time = &c->times[BCH_TIME_journal_noflush_write];
743 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
744 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
746 bch2_fs_btree_cache_init_early(&c->btree_cache);
748 mutex_init(&c->sectors_available_lock);
750 ret = percpu_init_rwsem(&c->mark_lock);
754 mutex_lock(&c->sb_lock);
755 ret = bch2_sb_to_fs(c, sb);
756 mutex_unlock(&c->sb_lock);
761 pr_uuid(&name, c->sb.user_uuid.b);
762 strscpy(c->name, name.buf, sizeof(c->name));
763 printbuf_exit(&name);
765 ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
770 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
771 !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
772 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
774 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
775 !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
776 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
778 c->opts = bch2_opts_default;
779 ret = bch2_opts_from_sb(&c->opts, sb);
783 bch2_opts_apply(&c->opts, opts);
785 c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
786 if (c->opts.inodes_use_key_cache)
787 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
789 c->block_bits = ilog2(block_sectors(c));
790 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
792 if (bch2_fs_init_fault("fs_alloc")) {
793 bch_err(c, "fs_alloc fault injected");
798 iter_size = sizeof(struct sort_iter) +
799 (btree_blocks(c) + 1) * 2 *
800 sizeof(struct sort_iter_set);
802 c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
804 if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
805 WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
806 !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
807 WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
808 !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
809 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
810 !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
811 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
812 !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
814 #ifndef BCH_WRITE_REF_DEBUG
815 percpu_ref_init(&c->writes, bch2_writes_disabled,
816 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
818 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
819 bioset_init(&c->btree_bio, 1,
820 max(offsetof(struct btree_read_bio, bio),
821 offsetof(struct btree_write_bio, wbio.bio)),
822 BIOSET_NEED_BVECS) ||
823 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
824 !(c->online_reserved = alloc_percpu(u64)) ||
825 !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
826 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
828 mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
829 !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
830 sizeof(u64), GFP_KERNEL))) {
831 ret = -BCH_ERR_ENOMEM_fs_other_alloc;
835 ret = bch2_fs_counters_init(c) ?:
836 bch2_io_clock_init(&c->io_clock[READ]) ?:
837 bch2_io_clock_init(&c->io_clock[WRITE]) ?:
838 bch2_fs_journal_init(&c->journal) ?:
839 bch2_fs_replicas_init(c) ?:
840 bch2_fs_btree_cache_init(c) ?:
841 bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
842 bch2_fs_btree_iter_init(c) ?:
843 bch2_fs_btree_interior_update_init(c) ?:
844 bch2_fs_buckets_waiting_for_journal_init(c) ?:
845 bch2_fs_btree_write_buffer_init(c) ?:
846 bch2_fs_subvolumes_init(c) ?:
847 bch2_fs_io_init(c) ?:
848 bch2_fs_nocow_locking_init(c) ?:
849 bch2_fs_encryption_init(c) ?:
850 bch2_fs_compress_init(c) ?:
851 bch2_fs_ec_init(c) ?:
852 bch2_fs_fsio_init(c) ?:
853 bch2_fs_fs_io_buffered_init(c);
854 bch2_fs_fs_io_direct_init(c);
858 mi = bch2_sb_get_members(c->disk_sb.sb);
859 for (i = 0; i < c->sb.nr_devices; i++)
860 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
861 bch2_dev_alloc(c, i)) {
866 bch2_journal_entry_res_resize(&c->journal,
867 &c->btree_root_journal_res,
868 BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
869 bch2_dev_usage_journal_reserve(c);
870 bch2_journal_entry_res_resize(&c->journal,
871 &c->clock_journal_res,
872 (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
874 mutex_lock(&bch_fs_list_lock);
875 ret = bch2_fs_online(c);
876 mutex_unlock(&bch_fs_list_lock);
889 static void print_mount_opts(struct bch_fs *c)
892 struct printbuf p = PRINTBUF;
895 prt_str(&p, "mounting version ");
896 bch2_version_to_text(&p, c->sb.version);
898 if (c->opts.read_only) {
899 prt_str(&p, " opts=");
901 prt_printf(&p, "ro");
904 for (i = 0; i < bch2_opts_nr; i++) {
905 const struct bch_option *opt = &bch2_opt_table[i];
906 u64 v = bch2_opt_get_by_id(&c->opts, i);
908 if (!(opt->flags & OPT_MOUNT))
911 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
914 prt_str(&p, first ? " opts=" : ",");
916 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
919 bch_info(c, "%s", p.buf);
923 int bch2_fs_start(struct bch_fs *c)
925 struct bch_sb_field_members *mi;
927 time64_t now = ktime_get_real_seconds();
933 down_write(&c->state_lock);
935 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
937 mutex_lock(&c->sb_lock);
939 for_each_online_member(ca, c, i)
940 bch2_sb_from_fs(c, ca);
942 mi = bch2_sb_get_members(c->disk_sb.sb);
943 for_each_online_member(ca, c, i)
944 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
946 mutex_unlock(&c->sb_lock);
948 for_each_rw_member(ca, c, i)
949 bch2_dev_allocator_add(c, ca);
950 bch2_recalc_capacity(c);
952 for (i = 0; i < BCH_TRANSACTIONS_NR; i++) {
953 mutex_lock(&c->btree_transaction_stats[i].lock);
954 bch2_time_stats_init(&c->btree_transaction_stats[i].lock_hold_times);
955 mutex_unlock(&c->btree_transaction_stats[i].lock);
958 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
959 ? bch2_fs_recovery(c)
960 : bch2_fs_initialize(c);
964 ret = bch2_opts_check_may_set(c);
968 if (bch2_fs_init_fault("fs_start")) {
969 bch_err(c, "fs_start fault injected");
974 set_bit(BCH_FS_STARTED, &c->flags);
976 if (c->opts.read_only || c->opts.nochanges) {
977 bch2_fs_read_only(c);
979 ret = !test_bit(BCH_FS_RW, &c->flags)
980 ? bch2_fs_read_write(c)
981 : bch2_fs_read_write_late(c);
988 up_write(&c->state_lock);
991 bch_err(c, "error starting filesystem: %s", bch2_err_str(ret));
995 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
997 struct bch_sb_field_members *sb_mi;
999 sb_mi = bch2_sb_get_members(sb);
1001 return -BCH_ERR_member_info_missing;
1003 if (le16_to_cpu(sb->block_size) != block_sectors(c))
1004 return -BCH_ERR_mismatched_block_size;
1006 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
1007 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1008 return -BCH_ERR_bucket_size_too_small;
1013 static int bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
1015 struct bch_sb *newest =
1016 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
1017 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
1019 if (!uuid_equal(&fs->uuid, &sb->uuid))
1020 return -BCH_ERR_device_not_a_member_of_filesystem;
1022 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
1023 return -BCH_ERR_device_has_been_removed;
1025 if (fs->block_size != sb->block_size)
1026 return -BCH_ERR_mismatched_block_size;
1031 /* Device startup/shutdown: */
1033 static void bch2_dev_release(struct kobject *kobj)
1035 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1040 static void bch2_dev_free(struct bch_dev *ca)
1042 cancel_work_sync(&ca->io_error_work);
1044 if (ca->kobj.state_in_sysfs &&
1046 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1048 if (ca->kobj.state_in_sysfs)
1049 kobject_del(&ca->kobj);
1051 bch2_free_super(&ca->disk_sb);
1052 bch2_dev_journal_exit(ca);
1054 free_percpu(ca->io_done);
1055 bioset_exit(&ca->replica_set);
1056 bch2_dev_buckets_free(ca);
1057 free_page((unsigned long) ca->sb_read_scratch);
1059 bch2_time_stats_exit(&ca->io_latency[WRITE]);
1060 bch2_time_stats_exit(&ca->io_latency[READ]);
1062 percpu_ref_exit(&ca->io_ref);
1063 percpu_ref_exit(&ca->ref);
1064 kobject_put(&ca->kobj);
1067 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1070 lockdep_assert_held(&c->state_lock);
1072 if (percpu_ref_is_zero(&ca->io_ref))
1075 __bch2_dev_read_only(c, ca);
1077 reinit_completion(&ca->io_ref_completion);
1078 percpu_ref_kill(&ca->io_ref);
1079 wait_for_completion(&ca->io_ref_completion);
1081 if (ca->kobj.state_in_sysfs) {
1082 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1083 sysfs_remove_link(&ca->kobj, "block");
1086 bch2_free_super(&ca->disk_sb);
1087 bch2_dev_journal_exit(ca);
1090 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1092 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1094 complete(&ca->ref_completion);
1097 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1099 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1101 complete(&ca->io_ref_completion);
1104 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1108 if (!c->kobj.state_in_sysfs)
1111 if (!ca->kobj.state_in_sysfs) {
1112 ret = kobject_add(&ca->kobj, &c->kobj,
1113 "dev-%u", ca->dev_idx);
1118 if (ca->disk_sb.bdev) {
1119 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1121 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1125 ret = sysfs_create_link(&ca->kobj, block, "block");
1133 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1134 struct bch_member *member)
1138 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1142 kobject_init(&ca->kobj, &bch2_dev_ktype);
1143 init_completion(&ca->ref_completion);
1144 init_completion(&ca->io_ref_completion);
1146 init_rwsem(&ca->bucket_lock);
1148 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1150 bch2_time_stats_init(&ca->io_latency[READ]);
1151 bch2_time_stats_init(&ca->io_latency[WRITE]);
1153 ca->mi = bch2_mi_to_cpu(member);
1154 ca->uuid = member->uuid;
1156 ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1157 ca->mi.bucket_size / btree_sectors(c));
1159 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1161 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1162 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1163 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1164 bch2_dev_buckets_alloc(c, ca) ||
1165 bioset_init(&ca->replica_set, 4,
1166 offsetof(struct bch_write_bio, bio), 0) ||
1167 !(ca->io_done = alloc_percpu(*ca->io_done)))
1176 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1179 ca->dev_idx = dev_idx;
1180 __set_bit(ca->dev_idx, ca->self.d);
1181 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1184 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1186 if (bch2_dev_sysfs_online(c, ca))
1187 pr_warn("error creating sysfs objects");
1190 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1192 struct bch_member *member =
1193 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1194 struct bch_dev *ca = NULL;
1197 if (bch2_fs_init_fault("dev_alloc"))
1200 ca = __bch2_dev_alloc(c, member);
1206 bch2_dev_attach(c, ca, dev_idx);
1211 return -BCH_ERR_ENOMEM_dev_alloc;
1214 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1218 if (bch2_dev_is_online(ca)) {
1219 bch_err(ca, "already have device online in slot %u",
1221 return -BCH_ERR_device_already_online;
1224 if (get_capacity(sb->bdev->bd_disk) <
1225 ca->mi.bucket_size * ca->mi.nbuckets) {
1226 bch_err(ca, "cannot online: device too small");
1227 return -BCH_ERR_device_size_too_small;
1230 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1232 ret = bch2_dev_journal_init(ca, sb->sb);
1238 if (sb->mode & FMODE_EXCL)
1239 ca->disk_sb.bdev->bd_holder = ca;
1240 memset(sb, 0, sizeof(*sb));
1242 ca->dev = ca->disk_sb.bdev->bd_dev;
1244 percpu_ref_reinit(&ca->io_ref);
1249 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1254 lockdep_assert_held(&c->state_lock);
1256 if (le64_to_cpu(sb->sb->seq) >
1257 le64_to_cpu(c->disk_sb.sb->seq))
1258 bch2_sb_to_fs(c, sb->sb);
1260 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1261 !c->devs[sb->sb->dev_idx]);
1263 ca = bch_dev_locked(c, sb->sb->dev_idx);
1265 ret = __bch2_dev_attach_bdev(ca, sb);
1269 bch2_dev_sysfs_online(c, ca);
1271 if (c->sb.nr_devices == 1)
1272 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1273 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1275 rebalance_wakeup(c);
1279 /* Device management: */
1282 * Note: this function is also used by the error paths - when a particular
1283 * device sees an error, we call it to determine whether we can just set the
1284 * device RO, or - if this function returns false - we'll set the whole
1287 * XXX: maybe we should be more explicit about whether we're changing state
1288 * because we got an error or what have you?
1290 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1291 enum bch_member_state new_state, int flags)
1293 struct bch_devs_mask new_online_devs;
1294 struct bch_dev *ca2;
1295 int i, nr_rw = 0, required;
1297 lockdep_assert_held(&c->state_lock);
1299 switch (new_state) {
1300 case BCH_MEMBER_STATE_rw:
1302 case BCH_MEMBER_STATE_ro:
1303 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1306 /* do we have enough devices to write to? */
1307 for_each_member_device(ca2, c, i)
1309 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1311 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1312 ? c->opts.metadata_replicas
1313 : c->opts.metadata_replicas_required,
1314 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1315 ? c->opts.data_replicas
1316 : c->opts.data_replicas_required);
1318 return nr_rw >= required;
1319 case BCH_MEMBER_STATE_failed:
1320 case BCH_MEMBER_STATE_spare:
1321 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1322 ca->mi.state != BCH_MEMBER_STATE_ro)
1325 /* do we have enough devices to read from? */
1326 new_online_devs = bch2_online_devs(c);
1327 __clear_bit(ca->dev_idx, new_online_devs.d);
1329 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1335 static bool bch2_fs_may_start(struct bch_fs *c)
1337 struct bch_sb_field_members *mi;
1339 unsigned i, flags = 0;
1341 if (c->opts.very_degraded)
1342 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1344 if (c->opts.degraded)
1345 flags |= BCH_FORCE_IF_DEGRADED;
1347 if (!c->opts.degraded &&
1348 !c->opts.very_degraded) {
1349 mutex_lock(&c->sb_lock);
1350 mi = bch2_sb_get_members(c->disk_sb.sb);
1352 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1353 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1356 ca = bch_dev_locked(c, i);
1358 if (!bch2_dev_is_online(ca) &&
1359 (ca->mi.state == BCH_MEMBER_STATE_rw ||
1360 ca->mi.state == BCH_MEMBER_STATE_ro)) {
1361 mutex_unlock(&c->sb_lock);
1365 mutex_unlock(&c->sb_lock);
1368 return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1371 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1374 * The allocator thread itself allocates btree nodes, so stop it first:
1376 bch2_dev_allocator_remove(c, ca);
1377 bch2_dev_journal_stop(&c->journal, ca);
1380 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1382 lockdep_assert_held(&c->state_lock);
1384 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1386 bch2_dev_allocator_add(c, ca);
1387 bch2_recalc_capacity(c);
1390 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1391 enum bch_member_state new_state, int flags)
1393 struct bch_sb_field_members *mi;
1396 if (ca->mi.state == new_state)
1399 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1400 return -BCH_ERR_device_state_not_allowed;
1402 if (new_state != BCH_MEMBER_STATE_rw)
1403 __bch2_dev_read_only(c, ca);
1405 bch_notice(ca, "%s", bch2_member_states[new_state]);
1407 mutex_lock(&c->sb_lock);
1408 mi = bch2_sb_get_members(c->disk_sb.sb);
1409 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1410 bch2_write_super(c);
1411 mutex_unlock(&c->sb_lock);
1413 if (new_state == BCH_MEMBER_STATE_rw)
1414 __bch2_dev_read_write(c, ca);
1416 rebalance_wakeup(c);
1421 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1422 enum bch_member_state new_state, int flags)
1426 down_write(&c->state_lock);
1427 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1428 up_write(&c->state_lock);
1433 /* Device add/removal: */
1435 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1437 struct bpos start = POS(ca->dev_idx, 0);
1438 struct bpos end = POS(ca->dev_idx, U64_MAX);
1442 * We clear the LRU and need_discard btrees first so that we don't race
1443 * with bch2_do_invalidates() and bch2_do_discards()
1445 ret = bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1446 BTREE_TRIGGER_NORUN, NULL) ?:
1447 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1448 BTREE_TRIGGER_NORUN, NULL) ?:
1449 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1450 BTREE_TRIGGER_NORUN, NULL) ?:
1451 bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1452 BTREE_TRIGGER_NORUN, NULL) ?:
1453 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1454 BTREE_TRIGGER_NORUN, NULL) ?:
1455 bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1456 BTREE_TRIGGER_NORUN, NULL);
1458 bch_err(c, "error removing dev alloc info: %s", bch2_err_str(ret));
1463 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1465 struct bch_sb_field_members *mi;
1466 unsigned dev_idx = ca->dev_idx, data;
1469 down_write(&c->state_lock);
1472 * We consume a reference to ca->ref, regardless of whether we succeed
1475 percpu_ref_put(&ca->ref);
1477 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1478 bch_err(ca, "Cannot remove without losing data");
1479 ret = -BCH_ERR_device_state_not_allowed;
1483 __bch2_dev_read_only(c, ca);
1485 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1487 bch_err(ca, "Remove failed: error dropping data: %s", bch2_err_str(ret));
1491 ret = bch2_dev_remove_alloc(c, ca);
1493 bch_err(ca, "Remove failed, error deleting alloc info");
1497 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1499 bch_err(ca, "Remove failed: error flushing journal: %s", bch2_err_str(ret));
1503 ret = bch2_journal_flush(&c->journal);
1505 bch_err(ca, "Remove failed, journal error");
1509 ret = bch2_replicas_gc2(c);
1511 bch_err(ca, "Remove failed: error from replicas gc: %s", bch2_err_str(ret));
1515 data = bch2_dev_has_data(c, ca);
1517 struct printbuf data_has = PRINTBUF;
1519 prt_bitflags(&data_has, bch2_data_types, data);
1520 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1521 printbuf_exit(&data_has);
1526 __bch2_dev_offline(c, ca);
1528 mutex_lock(&c->sb_lock);
1529 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1530 mutex_unlock(&c->sb_lock);
1532 percpu_ref_kill(&ca->ref);
1533 wait_for_completion(&ca->ref_completion);
1538 * At this point the device object has been removed in-core, but the
1539 * on-disk journal might still refer to the device index via sb device
1540 * usage entries. Recovery fails if it sees usage information for an
1541 * invalid device. Flush journal pins to push the back of the journal
1542 * past now invalid device index references before we update the
1543 * superblock, but after the device object has been removed so any
1544 * further journal writes elide usage info for the device.
1546 bch2_journal_flush_all_pins(&c->journal);
1549 * Free this device's slot in the bch_member array - all pointers to
1550 * this device must be gone:
1552 mutex_lock(&c->sb_lock);
1553 mi = bch2_sb_get_members(c->disk_sb.sb);
1554 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1556 bch2_write_super(c);
1558 mutex_unlock(&c->sb_lock);
1559 up_write(&c->state_lock);
1561 bch2_dev_usage_journal_reserve(c);
1564 if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1565 !percpu_ref_is_zero(&ca->io_ref))
1566 __bch2_dev_read_write(c, ca);
1567 up_write(&c->state_lock);
1571 /* Add new device to running filesystem: */
1572 int bch2_dev_add(struct bch_fs *c, const char *path)
1574 struct bch_opts opts = bch2_opts_empty();
1575 struct bch_sb_handle sb;
1576 struct bch_dev *ca = NULL;
1577 struct bch_sb_field_members *mi;
1578 struct bch_member dev_mi;
1579 unsigned dev_idx, nr_devices, u64s;
1580 struct printbuf errbuf = PRINTBUF;
1581 struct printbuf label = PRINTBUF;
1584 ret = bch2_read_super(path, &opts, &sb);
1586 bch_err(c, "device add error: error reading super: %s", bch2_err_str(ret));
1590 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1592 if (BCH_MEMBER_GROUP(&dev_mi)) {
1593 bch2_disk_path_to_text(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1594 if (label.allocation_failure) {
1600 ret = bch2_dev_may_add(sb.sb, c);
1602 bch_err(c, "device add error: %s", bch2_err_str(ret));
1606 ca = __bch2_dev_alloc(c, &dev_mi);
1608 bch2_free_super(&sb);
1613 bch2_dev_usage_init(ca);
1615 ret = __bch2_dev_attach_bdev(ca, &sb);
1621 ret = bch2_dev_journal_alloc(ca);
1623 bch_err(c, "device add error: journal alloc failed");
1627 down_write(&c->state_lock);
1628 mutex_lock(&c->sb_lock);
1630 ret = bch2_sb_from_fs(c, ca);
1632 bch_err(c, "device add error: new device superblock too small");
1636 mi = bch2_sb_get_members(ca->disk_sb.sb);
1638 if (!bch2_sb_resize_members(&ca->disk_sb,
1639 le32_to_cpu(mi->field.u64s) +
1640 sizeof(dev_mi) / sizeof(u64))) {
1641 bch_err(c, "device add error: new device superblock too small");
1642 ret = -BCH_ERR_ENOSPC_sb_members;
1646 if (dynamic_fault("bcachefs:add:no_slot"))
1649 mi = bch2_sb_get_members(c->disk_sb.sb);
1650 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1651 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1654 bch_err(c, "device add error: already have maximum number of devices");
1655 ret = -BCH_ERR_ENOSPC_sb_members;
1659 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1660 u64s = (sizeof(struct bch_sb_field_members) +
1661 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1663 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1665 bch_err(c, "device add error: no room in superblock for member info");
1666 ret = -BCH_ERR_ENOSPC_sb_members;
1672 mi->members[dev_idx] = dev_mi;
1673 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1674 c->disk_sb.sb->nr_devices = nr_devices;
1676 ca->disk_sb.sb->dev_idx = dev_idx;
1677 bch2_dev_attach(c, ca, dev_idx);
1679 if (BCH_MEMBER_GROUP(&dev_mi)) {
1680 ret = __bch2_dev_group_set(c, ca, label.buf);
1682 bch_err(c, "device add error: error setting label");
1687 bch2_write_super(c);
1688 mutex_unlock(&c->sb_lock);
1690 bch2_dev_usage_journal_reserve(c);
1692 ret = bch2_trans_mark_dev_sb(c, ca);
1694 bch_err(c, "device add error: error marking new superblock: %s", bch2_err_str(ret));
1698 ret = bch2_fs_freespace_init(c);
1700 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1704 ca->new_fs_bucket_idx = 0;
1706 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1707 __bch2_dev_read_write(c, ca);
1709 up_write(&c->state_lock);
1713 mutex_unlock(&c->sb_lock);
1714 up_write(&c->state_lock);
1718 bch2_free_super(&sb);
1719 printbuf_exit(&label);
1720 printbuf_exit(&errbuf);
1723 up_write(&c->state_lock);
1728 /* Hot add existing device to running filesystem: */
1729 int bch2_dev_online(struct bch_fs *c, const char *path)
1731 struct bch_opts opts = bch2_opts_empty();
1732 struct bch_sb_handle sb = { NULL };
1733 struct bch_sb_field_members *mi;
1738 down_write(&c->state_lock);
1740 ret = bch2_read_super(path, &opts, &sb);
1742 up_write(&c->state_lock);
1746 dev_idx = sb.sb->dev_idx;
1748 ret = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1750 bch_err(c, "error bringing %s online: %s", path, bch2_err_str(ret));
1754 ret = bch2_dev_attach_bdev(c, &sb);
1758 ca = bch_dev_locked(c, dev_idx);
1760 ret = bch2_trans_mark_dev_sb(c, ca);
1762 bch_err(c, "error bringing %s online: error from bch2_trans_mark_dev_sb: %s",
1763 path, bch2_err_str(ret));
1767 if (ca->mi.state == BCH_MEMBER_STATE_rw)
1768 __bch2_dev_read_write(c, ca);
1770 mutex_lock(&c->sb_lock);
1771 mi = bch2_sb_get_members(c->disk_sb.sb);
1773 mi->members[ca->dev_idx].last_mount =
1774 cpu_to_le64(ktime_get_real_seconds());
1776 bch2_write_super(c);
1777 mutex_unlock(&c->sb_lock);
1779 ret = bch2_fs_freespace_init(c);
1781 bch_err(c, "device add error: error initializing free space: %s", bch2_err_str(ret));
1783 up_write(&c->state_lock);
1786 up_write(&c->state_lock);
1787 bch2_free_super(&sb);
1791 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1793 down_write(&c->state_lock);
1795 if (!bch2_dev_is_online(ca)) {
1796 bch_err(ca, "Already offline");
1797 up_write(&c->state_lock);
1801 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1802 bch_err(ca, "Cannot offline required disk");
1803 up_write(&c->state_lock);
1804 return -BCH_ERR_device_state_not_allowed;
1807 __bch2_dev_offline(c, ca);
1809 up_write(&c->state_lock);
1813 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1815 struct bch_member *mi;
1818 down_write(&c->state_lock);
1820 if (nbuckets < ca->mi.nbuckets) {
1821 bch_err(ca, "Cannot shrink yet");
1826 if (bch2_dev_is_online(ca) &&
1827 get_capacity(ca->disk_sb.bdev->bd_disk) <
1828 ca->mi.bucket_size * nbuckets) {
1829 bch_err(ca, "New size larger than device");
1830 ret = -BCH_ERR_device_size_too_small;
1834 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1836 bch_err(ca, "Resize error: %s", bch2_err_str(ret));
1840 ret = bch2_trans_mark_dev_sb(c, ca);
1844 mutex_lock(&c->sb_lock);
1845 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1846 mi->nbuckets = cpu_to_le64(nbuckets);
1848 bch2_write_super(c);
1849 mutex_unlock(&c->sb_lock);
1851 bch2_recalc_capacity(c);
1853 up_write(&c->state_lock);
1857 /* return with ref on ca->ref: */
1858 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1864 for_each_member_device_rcu(ca, c, i, NULL)
1865 if (!strcmp(name, ca->name))
1867 ca = ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1874 /* Filesystem open: */
1876 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1877 struct bch_opts opts)
1879 struct bch_sb_handle *sb = NULL;
1880 struct bch_fs *c = NULL;
1881 struct bch_sb_field_members *mi;
1882 unsigned i, best_sb = 0;
1883 struct printbuf errbuf = PRINTBUF;
1886 if (!try_module_get(THIS_MODULE))
1887 return ERR_PTR(-ENODEV);
1894 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1900 for (i = 0; i < nr_devices; i++) {
1901 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1907 for (i = 1; i < nr_devices; i++)
1908 if (le64_to_cpu(sb[i].sb->seq) >
1909 le64_to_cpu(sb[best_sb].sb->seq))
1912 mi = bch2_sb_get_members(sb[best_sb].sb);
1915 while (i < nr_devices) {
1917 !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1918 pr_info("%pg has been removed, skipping", sb[i].bdev);
1919 bch2_free_super(&sb[i]);
1920 array_remove_item(sb, nr_devices, i);
1924 ret = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1930 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1936 down_write(&c->state_lock);
1937 for (i = 0; i < nr_devices; i++) {
1938 ret = bch2_dev_attach_bdev(c, &sb[i]);
1940 up_write(&c->state_lock);
1944 up_write(&c->state_lock);
1946 if (!bch2_fs_may_start(c)) {
1947 ret = -BCH_ERR_insufficient_devices_to_start;
1951 if (!c->opts.nostart) {
1952 ret = bch2_fs_start(c);
1958 printbuf_exit(&errbuf);
1959 module_put(THIS_MODULE);
1962 pr_err("bch_fs_open err opening %s: %s",
1963 devices[0], bch2_err_str(ret));
1965 if (!IS_ERR_OR_NULL(c))
1968 for (i = 0; i < nr_devices; i++)
1969 bch2_free_super(&sb[i]);
1974 /* Global interfaces/init */
1976 static void bcachefs_exit(void)
1980 bch2_chardev_exit();
1981 bch2_btree_key_cache_exit();
1983 kset_unregister(bcachefs_kset);
1986 static int __init bcachefs_init(void)
1988 bch2_bkey_pack_test();
1990 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1991 bch2_btree_key_cache_init() ||
1992 bch2_chardev_init() ||
2003 #define BCH_DEBUG_PARAM(name, description) \
2005 module_param_named(name, bch2_##name, bool, 0644); \
2006 MODULE_PARM_DESC(name, description);
2008 #undef BCH_DEBUG_PARAM
2011 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2012 module_param_named(version, bch2_metadata_version, uint, 0400);
2014 module_exit(bcachefs_exit);
2015 module_init(bcachefs_init);